The pain — unique to this desk
Governance specialist · Governance HOLD
AI Ethics Analyst
Ethics as a slide after ship
What this agent actually does
Policy & auditability.
Ethics as a slide after ship. A SHA-256 pack treated as a letter. The losing claim erased because it lost the vote.
The need this desk closes
Safety goldens as a merge gate. Dissent kept. Art. 50 is a desk. The pack is hashed — the pack is not a letter.
Why a generic chat cannot fake this
Ethics as a slide after ship
Safety goldens as a merge gate. Dissent kept. Art. 50 is a desk. The pack is hashed — the pack is not a letter.
Takes
- Policy doc + samples
Returns
- Pass rate
- Dissent log
- Compliance gate list
Modalities this desk actually handles
How this specialist fuses
Ethics as a slide after ship
Policy goldens × refusal quality. Term-distance mapper is hash-ngram cosine — not a bias certificate.
Vision-language on this desk
Named family. Never assumed.
Vision-language bias is scored only if image goldens exist. A cosine on captions is not a fairness letter.
Patch / tokenize → project → fuse → ground is the host mechanism. This desk fuses late under its own playbook — not a slogan copied across the roster. Host mechanism.
A ship decision or a refusal is the ticket. caps.golden is the policy set. eval.run_gate holds merge if safety or jury fails. Dissent Log keeps the losing claim — outvoted is not erased. A SHA-256 pack is not a notified-body letter. Term-distance mapper is pairwise cosine on hash-ngrams — not a bias certificate. Compliance fix-all covers audit trail · RTBF · sovereignty chains as checklists, not as certifications.
Live missions run in Role Studio. Analog kernels stay in this tab until a hosted call is chosen. Authors never grade themselves. A hashed pack is not a letter. Hire bands are market replacement cost, not EngOS payroll. Isolation is probed, not assumed.
Reports up as Governance HOLD. Second family ai-trainer-evaluator. Authors never grade themselves. Isolation is probed, not assumed.
Five analog ticket kinds
How a hired human spends the week. What the analog kernel closes.
Percent splits stay human-week language. Never GPU-loop quotas. Each kind binds to the analog tool that closes it. Escalate only when the kernel cannot.
- AI Ethics Analyst analog closeHuman bar: Name the ticket. Attach evidence. Second family grades. Analog: #align. Kernel #/align on this host. Arithmetic / Canvas 2D. Device none. Escalate when: The analog kernel cannot close, or a hosted connector is required as if it were present.
A typical ticket
What you say. What sits.
Audit last week's ship decisions against the policy doc. Hold merge if the refusal golden fails. Keep the dissent. Do not call the pack a letter.
Sit keys ethics / Art. 50 / bias / dissent / disclosure. Ticket kinds eval and feature. Security shares injection/exfil. This desk owns the policy trail.
Playbook
How this agent works the ticket.
- 01 Policy goldensSafety and refusal cases. Pass rate is the artifact.
- 02 GateHold merge if safety or jury fails. Ethics is not a slide after ship.
- 03 DissentLosing claim preserved. Outvoted is not deleted.
- 04 TraceEvery ship decision tagged. Compliance can ask why.
- 05 Art. 50 deskDisclosure work has a place. The pack is hashed. The pack is not a letter.
- 06 Distance analogPairwise cosine on hash-ngrams. Not a bias certificate.
Acts like the role
Work it does. Work it will not fake.
Does
- Expand policy goldens
- Review refusals for quality, not only for rate
- Trace a ship decision
- Run compliance fix-all as a checklist
- Keep dissent
Does not
- Print a bias certificate from a cosine
- Treat a SHA-256 pack as a notified-body letter
- Erase the losing claim
- Let ethics wait until after GA
Jobs on the board
Scenarios this desk was built to close.
From Role Studio's scenario bank — current pain without Helix, and the job the agent actually runs. Top eight of twenty.
P1
Decision audit trail
Today. Cannot explain model decisions to compliance
This agent. Trace + policy tags on every ship decision
P0
Golden dataset regression
Today. Prompt/model change ships without suite
This agent. Versioned golden set + offline gate before merge
P0
Direct + indirect prompt injection
Today. OWASP #1 — injection via user and retrieved docs
This agent. Red-team pack + input/tool/output layers
P0
Secret / PII exfiltration
Today. Models echo keys from context
This agent. Secret scan on context + artifact write
P1
Success metrics on AI features
Today. Ship AI without measurable outcomes
This agent. Feature brief + eval plan + KPI log
P0
Claim-level citations
Today. Only 50% of sentences supported in audits
This agent. Per-claim support flags + hold if unsupported
P0
False completion graders
Today. Agent claims done while tests fail
This agent. Objective graders before done
P0
Prompt git-style versioning
Today. Prompts live in Notion / chat history
This agent. Registry with A/B + golden score
Skill.md
Policy review
Durable analog execution standard. Trigger, five ticket kinds, analog tools, must / must-not, handoff (evidence not authority), spend.halt, second family. Same factory as PR Creator, Code Reviewer, Handoff Verifier — plus fromDesk(ai-ethics-analyst).
Trigger. Agent decisions need a fairness and policy audit.
- 01 StepTake the policy doc + samples
- 02 StepRun safety goldens
- 03 StepHold merge on failure
- 04 StepKeep dissent
- 05 StepList gaps — do not certify
Must
- Audit trail
- Gaps listed
- Dissent preserved
Must not
- Call the pack a letter
- Issue a bias certificate from the analog
# AI Ethics Analyst id: ai-ethics-analyst layer: governance (Governance HOLD) second family: ai-trainer-evaluator tools: #/align ## Pain The hired AI Ethics Analyst seat does not exist yet, or the work has no named ticket. ## Need A named ticket, an analog close, and a second family. Not a chat that grades itself. ## Isolation Isolation is probed, not assumed. Unprobed stays unlabeled. Never certified-green from a desk. ## spend.halt spend.halt on the ticket cap. Only the operator raises the ceiling. ## Ticket kinds - align-pass · AI Ethics Analyst analog close · analog #/align · escalate when: The analog kernel cannot close, or a hosted connector is required as if it were present.
--- name: analog-desk isolation: probed pack_is_letter: false arr_claimed: false --- # Analog desk skill Authors never grade themselves. Second family sits the gate. spend.halt is the ceiling. Isolation stays unlabeled until probed.
Who sits with this desk
Swarm compose by ticket kind.
Not named on a ticket-kind composition. Talk Route can still sit it from sit keys.
Sit keys
Talk Route matches these words.
Keyword analog in this tab. Not a hosted model. Ticket id is the idempotency key.
RBAC
Pass bar 78. C1.
- Ceiling. C1 propose and gated read. Deploy and credential reveal denied.
- Deny. credentials.reveal · deploy.canary
- Scopes. workspace_read
- Swarm seats. Not named on a ticket-kind composition. Talk Route can still sit it.
Daily missions
Run against live engines. Not slides.
eval.run_gate
Run safety + jury gate
Hold merge if safety/jury fails
Accept: Gate decision explained. Pain removed: Ethics isn't a slide deck after ship. Surfaces: Eval Gate.
Superpowers
- Safety eval method
- Audit via event bus
- STE plain language
Daily jobs
- Expand policy goldens
- Review refusals
- Trace a ship decision
- Run compliance fix-all chains
Skills
- Governance
- Bias testing
- Compliance
- Policy
Toolkit
Command Center becomes this desk.
Eval Gate
Policy golden set
Safety/refusal cases
Outcome: Pass rate · caps.golden.
Command Center
Safety red-team
Injection & exfil probes
Outcome: Blocks · caps.redteam.
Kernel Guard
Compliance fix-all
Audit trail · RTBF · sovereignty chains
Outcome: Compliance gate list · findings.fix_all.
Work templates
eval
Policy review
Audit agent decisions for fairness and policy adherence
Paste: Policy doc + samples. Accept: Audit trail · Gaps listed.
Surfaces
Nav this desk actually opens.
Local analog
Term distance mapper
Pairwise cosine on hash-ngram vectors. Not a bias certification.
Term distance mapper: pairwise cosine on hash-ngram vectors. Not a bias certification. Dissent Log keeps the loser. SAMPLE.
Replacement cost
$110,000–$165,000 cash
Year-1 loaded + recruiting $203,500. Policy / Art. 50 desk. Below engineering ceiling.. Not EngOS payroll. Not ARR.
Pilot $0 / Team $79 sits this analog desk. Year-1 hire is $203,500 loaded. That is not a replacement claim. The hired role remains the real thing. The analog desk reports up so one operator can run the ticket.
Governance neighbors
Governance specialist
AI Security Specialist
Firewall claim on a regex
A firewall claim on a regex. Isolation painted certified without a probe. Secrets in a trace that left the tab.
Six-layer guard with receipts. Isolation probed, not assumed. A pattern hit is not a firewall.
- Red-team the support agent and keep the receipts
- Review the tool deny-list
- Sandbox-tier review for a new agent
Governance specialist
AI Trainer and Evaluator
The author grading the agent they wrote
The author grading the agent they wrote. A courtesy pass. The outvoted score deleted.
Independent 11-method gate. Pass bar does not go down. False-complete hunt. Dissent preserved.
- Grow the golden set
- Score prompt A/B
- Investigate false completes